Raw LLM Responses

AI is expensive to train, not run. If you have a consumer level graphics card (3080 or better, I reckon), you can run a decent quant of DeepSeek or Llama. You can utilize methodologies to make those self-hosted models punch WAY above their weight class. Things like RAG. There are very high-quality open source models you can run. Even training or tailoring models isn't really that expensive anymore, either. AI has basically no moat. As a developer, I actually think what we are calling AI is actually incredible in terms of how it works. It is, essentially, a very high dimensional search engine, which has a huge number of applications that have yet to really be realized. I do think companies are massively overstating the impact of their model, and really, are selling investors on the idea they are on the verge of replacing all labor, a capitalist pipe dream delusion. Mainstream AI development has really plateaud since the "all you need is attention" paper that really spurred on stuff like chatgpt, but there are a ton of use-cases. I've found several excellent uses for it, and im an idiot. The investments we see at fb and other companies, I believe, are less about the actual financial cost of running or training something like chatgpt or Facebook's models, but actually, what they are doing is trying to train it on basically everything they have about you, and essentially everyone and everything on the planet. That isnt necessary for something like chatgpt, but it is necessary to make a draconian minority report style hellscape, which fits Musk and Zucc and Altman. I think that is the true goals of their massive infrastructure deployments. We won't have access to those tools, i think. Maybe some derivative work, but their goals are basically to do what chatgpt does with words, but with everything you personally have ever said and done. That is a social media companies asset... data... and this is a tool that can construct a very high dimensional graph of relation